Construction of a machine learning-based screening model for IgD myeloma

Objective: Immunoglobulin D (IgD) myeloma is a rare subtype of multiple myeloma (MM), comprising approximately 1 %-2 % of all MM cases. Owing to the diminished levels of IgD in serum, IgD MM manifests as subtle M protein spikes in routine serum electrophoresis, rendering it susceptible to misdiagnosis and underdiagnosis. The objective of this study was to develop a machine learning (ML) model utilizing readily available complete blood count and biochemical test data for the purpose of screening IgD MM. Methods: This study encompassed clinical data from 83 newly diagnosed IgD MM patients and 166 non-IgD MM patients, decision tree, random forest, support vector machine (SVM), stochastic gradient descent (SGD), and adaptive boosting (AdaBoost) algorithms were employed for model construction. The predictive performance of the ML models was evaluated using the area under the receiver operating characteristic curve (AUC), calibration curve analysis, and decision curve analysis. Results: The random forest-based screening model demonstrated superior performance, incorporating seven key features: LDH, albumin, creatinine, Ca, beta 2 microglobulin, age and Hb. It achieved an AUC of 0.954 (95 % CI 0.930-0.977), with a sensitivity of 0.958, specificity of 0.747, positive predictive value of 88.3 % and negative predictive value of 89.9 %. Furthermore, this model has been evaluated in the validation cohort. Conclusions: The model constructed based on the random forest algorithm demonstrates potential in screening IgD MM patients, particularly when routine IgD immunotyping testing is not conducted in clinical practice. This can assist clinicians in early diagnosis and personalized treatment strategies, thereby optimizing the utilization of medical resources.